xref: /freebsd/sys/amd64/vmm/io/ppt.c (revision 413197cef9ccbf22831eb914c2ef469880468d97)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 NetApp, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/kernel.h>
32 #include <sys/malloc.h>
33 #include <sys/module.h>
34 #include <sys/bus.h>
35 #include <sys/pciio.h>
36 #include <sys/rman.h>
37 #include <sys/smp.h>
38 #include <sys/sysctl.h>
39 
40 #include <dev/pci/pcivar.h>
41 #include <dev/pci/pcireg.h>
42 
43 #include <machine/resource.h>
44 #include <machine/vmm.h>
45 #include <machine/vmm_dev.h>
46 
47 #include <dev/vmm/vmm_ktr.h>
48 
49 #include "vmm_lapic.h"
50 
51 #include "iommu.h"
52 #include "ppt.h"
53 
54 /* XXX locking */
55 
56 #define	MAX_MSIMSGS	32
57 
58 /*
59  * If the MSI-X table is located in the middle of a BAR then that MMIO
60  * region gets split into two segments - one segment above the MSI-X table
61  * and the other segment below the MSI-X table - with a hole in place of
62  * the MSI-X table so accesses to it can be trapped and emulated.
63  *
64  * So, allocate a MMIO segment for each BAR register + 1 additional segment.
65  */
66 #define	MAX_MMIOSEGS	((PCIR_MAX_BAR_0 + 1) + 1)
67 
68 MALLOC_DEFINE(M_PPTMSIX, "pptmsix", "Passthru MSI-X resources");
69 
70 struct pptintr_arg {				/* pptintr(pptintr_arg) */
71 	struct pptdev	*pptdev;
72 	uint64_t	addr;
73 	uint64_t	msg_data;
74 };
75 
76 struct pptseg {
77 	vm_paddr_t	gpa;
78 	size_t		len;
79 	int		wired;
80 };
81 
82 struct pptdev {
83 	device_t	dev;
84 	struct vm	*vm;			/* owner of this device */
85 	TAILQ_ENTRY(pptdev)	next;
86 	struct pptseg mmio[MAX_MMIOSEGS];
87 	struct {
88 		int	num_msgs;		/* guest state */
89 
90 		int	startrid;		/* host state */
91 		struct resource *res[MAX_MSIMSGS];
92 		void	*cookie[MAX_MSIMSGS];
93 		struct pptintr_arg arg[MAX_MSIMSGS];
94 	} msi;
95 
96 	struct {
97 		int num_msgs;
98 		int startrid;
99 		int msix_table_rid;
100 		int msix_pba_rid;
101 		struct resource *msix_table_res;
102 		struct resource *msix_pba_res;
103 		struct resource **res;
104 		void **cookie;
105 		struct pptintr_arg *arg;
106 	} msix;
107 };
108 
109 SYSCTL_DECL(_hw_vmm);
110 SYSCTL_NODE(_hw_vmm, OID_AUTO, ppt, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
111     "bhyve passthru devices");
112 
113 static int num_pptdevs;
114 SYSCTL_INT(_hw_vmm_ppt, OID_AUTO, devices, CTLFLAG_RD, &num_pptdevs, 0,
115     "number of pci passthru devices");
116 
117 static TAILQ_HEAD(, pptdev) pptdev_list = TAILQ_HEAD_INITIALIZER(pptdev_list);
118 
119 static int
ppt_probe(device_t dev)120 ppt_probe(device_t dev)
121 {
122 	int bus, slot, func;
123 	struct pci_devinfo *dinfo;
124 
125 	dinfo = (struct pci_devinfo *)device_get_ivars(dev);
126 
127 	bus = pci_get_bus(dev);
128 	slot = pci_get_slot(dev);
129 	func = pci_get_function(dev);
130 
131 	/*
132 	 * To qualify as a pci passthrough device a device must:
133 	 * - be allowed by administrator to be used in this role
134 	 * - be an endpoint device
135 	 */
136 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
137 		return (ENXIO);
138 	else if (vmm_is_pptdev(bus, slot, func))
139 		return (0);
140 	else
141 		/*
142 		 * Returning BUS_PROBE_NOWILDCARD here matches devices that the
143 		 * SR-IOV infrastructure specified as "ppt" passthrough devices.
144 		 * All normal devices that did not have "ppt" specified as their
145 		 * driver will not be matched by this.
146 		 */
147 		return (BUS_PROBE_NOWILDCARD);
148 }
149 
150 static int
ppt_attach(device_t dev)151 ppt_attach(device_t dev)
152 {
153 	struct pptdev *ppt;
154 	uint16_t cmd, cmd1;
155 	int error;
156 
157 	ppt = device_get_softc(dev);
158 
159 	cmd1 = cmd = pci_read_config(dev, PCIR_COMMAND, 2);
160 	cmd &= ~(PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
161 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
162 	error = iommu_remove_device(iommu_host_domain(), dev, pci_get_rid(dev));
163 	if (error != 0) {
164 		pci_write_config(dev, PCIR_COMMAND, cmd1, 2);
165 		return (error);
166 	}
167 	num_pptdevs++;
168 	TAILQ_INSERT_TAIL(&pptdev_list, ppt, next);
169 	ppt->dev = dev;
170 
171 	if (bootverbose)
172 		device_printf(dev, "attached\n");
173 
174 	return (0);
175 }
176 
177 static int
ppt_detach(device_t dev)178 ppt_detach(device_t dev)
179 {
180 	struct pptdev *ppt;
181 	int error;
182 
183 	ppt = device_get_softc(dev);
184 
185 	if (ppt->vm != NULL)
186 		return (EBUSY);
187 	if (iommu_host_domain() != NULL) {
188 		error = iommu_add_device(iommu_host_domain(), dev,
189 		    pci_get_rid(dev));
190 	} else {
191 		error = 0;
192 	}
193 	if (error != 0)
194 		return (error);
195 	num_pptdevs--;
196 	TAILQ_REMOVE(&pptdev_list, ppt, next);
197 
198 	return (0);
199 }
200 
201 static device_method_t ppt_methods[] = {
202 	/* Device interface */
203 	DEVMETHOD(device_probe,		ppt_probe),
204 	DEVMETHOD(device_attach,	ppt_attach),
205 	DEVMETHOD(device_detach,	ppt_detach),
206 	{0, 0}
207 };
208 
209 DEFINE_CLASS_0(ppt, ppt_driver, ppt_methods, sizeof(struct pptdev));
210 DRIVER_MODULE(ppt, pci, ppt_driver, NULL, NULL);
211 
212 static int
ppt_find(struct vm * vm,int bus,int slot,int func,struct pptdev ** pptp)213 ppt_find(struct vm *vm, int bus, int slot, int func, struct pptdev **pptp)
214 {
215 	device_t dev;
216 	struct pptdev *ppt;
217 	int b, s, f;
218 
219 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
220 		dev = ppt->dev;
221 		b = pci_get_bus(dev);
222 		s = pci_get_slot(dev);
223 		f = pci_get_function(dev);
224 		if (bus == b && slot == s && func == f)
225 			break;
226 	}
227 
228 	if (ppt == NULL)
229 		return (ENOENT);
230 	if (ppt->vm != vm)		/* Make sure we own this device */
231 		return (EBUSY);
232 	*pptp = ppt;
233 	return (0);
234 }
235 
236 static void
ppt_unmap_all_mmio(struct vm * vm,struct pptdev * ppt)237 ppt_unmap_all_mmio(struct vm *vm, struct pptdev *ppt)
238 {
239 	int i;
240 	struct pptseg *seg;
241 
242 	for (i = 0; i < MAX_MMIOSEGS; i++) {
243 		seg = &ppt->mmio[i];
244 		if (seg->len == 0)
245 			continue;
246 		(void)vm_unmap_mmio(vm, seg->gpa, seg->len);
247 		bzero(seg, sizeof(struct pptseg));
248 	}
249 }
250 
251 static void
ppt_teardown_msi(struct pptdev * ppt)252 ppt_teardown_msi(struct pptdev *ppt)
253 {
254 	int i, rid;
255 	void *cookie;
256 	struct resource *res;
257 
258 	if (ppt->msi.num_msgs == 0)
259 		return;
260 
261 	for (i = 0; i < ppt->msi.num_msgs; i++) {
262 		rid = ppt->msi.startrid + i;
263 		res = ppt->msi.res[i];
264 		cookie = ppt->msi.cookie[i];
265 
266 		if (cookie != NULL)
267 			bus_teardown_intr(ppt->dev, res, cookie);
268 
269 		if (res != NULL)
270 			bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res);
271 
272 		ppt->msi.res[i] = NULL;
273 		ppt->msi.cookie[i] = NULL;
274 	}
275 
276 	if (ppt->msi.startrid == 1)
277 		pci_release_msi(ppt->dev);
278 
279 	ppt->msi.num_msgs = 0;
280 }
281 
282 static void
ppt_teardown_msix_intr(struct pptdev * ppt,int idx)283 ppt_teardown_msix_intr(struct pptdev *ppt, int idx)
284 {
285 	int rid;
286 	struct resource *res;
287 	void *cookie;
288 
289 	rid = ppt->msix.startrid + idx;
290 	res = ppt->msix.res[idx];
291 	cookie = ppt->msix.cookie[idx];
292 
293 	if (cookie != NULL)
294 		bus_teardown_intr(ppt->dev, res, cookie);
295 
296 	if (res != NULL)
297 		bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res);
298 
299 	ppt->msix.res[idx] = NULL;
300 	ppt->msix.cookie[idx] = NULL;
301 }
302 
303 static void
ppt_teardown_msix(struct pptdev * ppt)304 ppt_teardown_msix(struct pptdev *ppt)
305 {
306 	int i;
307 
308 	if (ppt->msix.num_msgs == 0)
309 		return;
310 
311 	for (i = 0; i < ppt->msix.num_msgs; i++)
312 		ppt_teardown_msix_intr(ppt, i);
313 
314 	free(ppt->msix.res, M_PPTMSIX);
315 	free(ppt->msix.cookie, M_PPTMSIX);
316 	free(ppt->msix.arg, M_PPTMSIX);
317 
318 	pci_release_msi(ppt->dev);
319 
320 	if (ppt->msix.msix_table_res) {
321 		bus_release_resource(ppt->dev, SYS_RES_MEMORY,
322 				     ppt->msix.msix_table_rid,
323 				     ppt->msix.msix_table_res);
324 		ppt->msix.msix_table_res = NULL;
325 		ppt->msix.msix_table_rid = 0;
326 	}
327 	if (ppt->msix.msix_pba_res) {
328 		bus_release_resource(ppt->dev, SYS_RES_MEMORY,
329 				     ppt->msix.msix_pba_rid,
330 				     ppt->msix.msix_pba_res);
331 		ppt->msix.msix_pba_res = NULL;
332 		ppt->msix.msix_pba_rid = 0;
333 	}
334 
335 	ppt->msix.num_msgs = 0;
336 }
337 
338 int
ppt_avail_devices(void)339 ppt_avail_devices(void)
340 {
341 
342 	return (num_pptdevs);
343 }
344 
345 int
ppt_assigned_devices(struct vm * vm)346 ppt_assigned_devices(struct vm *vm)
347 {
348 	struct pptdev *ppt;
349 	int num;
350 
351 	num = 0;
352 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
353 		if (ppt->vm == vm)
354 			num++;
355 	}
356 	return (num);
357 }
358 
359 bool
ppt_is_mmio(struct vm * vm,vm_paddr_t gpa)360 ppt_is_mmio(struct vm *vm, vm_paddr_t gpa)
361 {
362 	int i;
363 	struct pptdev *ppt;
364 	struct pptseg *seg;
365 
366 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
367 		if (ppt->vm != vm)
368 			continue;
369 
370 		for (i = 0; i < MAX_MMIOSEGS; i++) {
371 			seg = &ppt->mmio[i];
372 			if (seg->len == 0)
373 				continue;
374 			if (gpa >= seg->gpa && gpa < seg->gpa + seg->len)
375 				return (true);
376 		}
377 	}
378 
379 	return (false);
380 }
381 
382 static void
ppt_pci_reset(device_t dev)383 ppt_pci_reset(device_t dev)
384 {
385 
386 	if (pcie_flr(dev,
387 	     max(pcie_get_max_completion_timeout(dev) / 1000, 10), true))
388 		return;
389 
390 	pci_power_reset(dev);
391 }
392 
393 static uint16_t
ppt_bar_enables(struct pptdev * ppt)394 ppt_bar_enables(struct pptdev *ppt)
395 {
396 	struct pci_map *pm;
397 	uint16_t cmd;
398 
399 	cmd = 0;
400 	for (pm = pci_first_bar(ppt->dev); pm != NULL; pm = pci_next_bar(pm)) {
401 		if (PCI_BAR_IO(pm->pm_value))
402 			cmd |= PCIM_CMD_PORTEN;
403 		if (PCI_BAR_MEM(pm->pm_value))
404 			cmd |= PCIM_CMD_MEMEN;
405 	}
406 	return (cmd);
407 }
408 
409 int
ppt_assign_device(struct vm * vm,int bus,int slot,int func)410 ppt_assign_device(struct vm *vm, int bus, int slot, int func)
411 {
412 	struct pptdev *ppt;
413 	int error;
414 	uint16_t cmd;
415 
416 	/* Passing NULL requires the device to be unowned. */
417 	error = ppt_find(NULL, bus, slot, func, &ppt);
418 	if (error)
419 		return (error);
420 
421 	pci_save_state(ppt->dev);
422 	ppt_pci_reset(ppt->dev);
423 	pci_restore_state(ppt->dev);
424 	error = iommu_add_device(vm_iommu_domain(vm), ppt->dev,
425 	    pci_get_rid(ppt->dev));
426 	if (error != 0)
427 		return (error);
428 	ppt->vm = vm;
429 	cmd = pci_read_config(ppt->dev, PCIR_COMMAND, 2);
430 	cmd |= PCIM_CMD_BUSMASTEREN | ppt_bar_enables(ppt);
431 	pci_write_config(ppt->dev, PCIR_COMMAND, cmd, 2);
432 	return (0);
433 }
434 
435 int
ppt_unassign_device(struct vm * vm,int bus,int slot,int func)436 ppt_unassign_device(struct vm *vm, int bus, int slot, int func)
437 {
438 	struct pptdev *ppt;
439 	int error;
440 	uint16_t cmd;
441 
442 	error = ppt_find(vm, bus, slot, func, &ppt);
443 	if (error)
444 		return (error);
445 
446 	cmd = pci_read_config(ppt->dev, PCIR_COMMAND, 2);
447 	cmd &= ~(PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
448 	pci_write_config(ppt->dev, PCIR_COMMAND, cmd, 2);
449 	pci_save_state(ppt->dev);
450 	ppt_pci_reset(ppt->dev);
451 	pci_restore_state(ppt->dev);
452 	ppt_unmap_all_mmio(vm, ppt);
453 	ppt_teardown_msi(ppt);
454 	ppt_teardown_msix(ppt);
455 	error = iommu_remove_device(vm_iommu_domain(vm), ppt->dev,
456 	    pci_get_rid(ppt->dev));
457 	ppt->vm = NULL;
458 	return (error);
459 }
460 
461 int
ppt_unassign_all(struct vm * vm)462 ppt_unassign_all(struct vm *vm)
463 {
464 	struct pptdev *ppt;
465 	int bus, slot, func;
466 	device_t dev;
467 
468 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
469 		if (ppt->vm == vm) {
470 			dev = ppt->dev;
471 			bus = pci_get_bus(dev);
472 			slot = pci_get_slot(dev);
473 			func = pci_get_function(dev);
474 			vm_unassign_pptdev(vm, bus, slot, func);
475 		}
476 	}
477 
478 	return (0);
479 }
480 
481 static bool
ppt_valid_bar_mapping(struct pptdev * ppt,vm_paddr_t hpa,size_t len)482 ppt_valid_bar_mapping(struct pptdev *ppt, vm_paddr_t hpa, size_t len)
483 {
484 	struct pci_map *pm;
485 	pci_addr_t base, size;
486 
487 	for (pm = pci_first_bar(ppt->dev); pm != NULL; pm = pci_next_bar(pm)) {
488 		if (!PCI_BAR_MEM(pm->pm_value))
489 			continue;
490 		base = pm->pm_value & PCIM_BAR_MEM_BASE;
491 		size = (pci_addr_t)1 << pm->pm_size;
492 		if (hpa >= base && hpa + len <= base + size)
493 			return (true);
494 	}
495 	return (false);
496 }
497 
498 int
ppt_map_mmio(struct vm * vm,int bus,int slot,int func,vm_paddr_t gpa,size_t len,vm_paddr_t hpa)499 ppt_map_mmio(struct vm *vm, int bus, int slot, int func,
500 	     vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
501 {
502 	int i, error;
503 	struct pptseg *seg;
504 	struct pptdev *ppt;
505 
506 	if (len % PAGE_SIZE != 0 || len == 0 || gpa % PAGE_SIZE != 0 ||
507 	    hpa % PAGE_SIZE != 0 || gpa + len < gpa || hpa + len < hpa)
508 		return (EINVAL);
509 
510 	error = ppt_find(vm, bus, slot, func, &ppt);
511 	if (error)
512 		return (error);
513 
514 	if (!ppt_valid_bar_mapping(ppt, hpa, len))
515 		return (EINVAL);
516 
517 	for (i = 0; i < MAX_MMIOSEGS; i++) {
518 		seg = &ppt->mmio[i];
519 		if (seg->len == 0) {
520 			error = vm_map_mmio(vm, gpa, len, hpa);
521 			if (error == 0) {
522 				seg->gpa = gpa;
523 				seg->len = len;
524 			}
525 			return (error);
526 		}
527 	}
528 	return (ENOSPC);
529 }
530 
531 int
ppt_unmap_mmio(struct vm * vm,int bus,int slot,int func,vm_paddr_t gpa,size_t len)532 ppt_unmap_mmio(struct vm *vm, int bus, int slot, int func,
533 	       vm_paddr_t gpa, size_t len)
534 {
535 	int i, error;
536 	struct pptseg *seg;
537 	struct pptdev *ppt;
538 
539 	error = ppt_find(vm, bus, slot, func, &ppt);
540 	if (error)
541 		return (error);
542 
543 	for (i = 0; i < MAX_MMIOSEGS; i++) {
544 		seg = &ppt->mmio[i];
545 		if (seg->gpa == gpa && seg->len == len) {
546 			error = vm_unmap_mmio(vm, seg->gpa, seg->len);
547 			if (error == 0) {
548 				seg->gpa = 0;
549 				seg->len = 0;
550 			}
551 			return (error);
552 		}
553 	}
554 	return (ENOENT);
555 }
556 
557 static int
pptintr(void * arg)558 pptintr(void *arg)
559 {
560 	struct pptdev *ppt;
561 	struct pptintr_arg *pptarg;
562 
563 	pptarg = arg;
564 	ppt = pptarg->pptdev;
565 
566 	if (ppt->vm != NULL)
567 		lapic_intr_msi(ppt->vm, pptarg->addr, pptarg->msg_data);
568 	else {
569 		/*
570 		 * XXX
571 		 * This is not expected to happen - panic?
572 		 */
573 	}
574 
575 	/*
576 	 * For legacy interrupts give other filters a chance in case
577 	 * the interrupt was not generated by the passthrough device.
578 	 */
579 	if (ppt->msi.startrid == 0)
580 		return (FILTER_STRAY);
581 	else
582 		return (FILTER_HANDLED);
583 }
584 
585 int
ppt_setup_msi(struct vm * vm,int bus,int slot,int func,uint64_t addr,uint64_t msg,int numvec)586 ppt_setup_msi(struct vm *vm, int bus, int slot, int func,
587 	      uint64_t addr, uint64_t msg, int numvec)
588 {
589 	int i, rid, flags;
590 	int msi_count, startrid, error, tmp;
591 	struct pptdev *ppt;
592 
593 	if (numvec < 0 || numvec > MAX_MSIMSGS)
594 		return (EINVAL);
595 
596 	error = ppt_find(vm, bus, slot, func, &ppt);
597 	if (error)
598 		return (error);
599 
600 	/* Reject attempts to enable MSI while MSI-X is active. */
601 	if (ppt->msix.num_msgs != 0 && numvec != 0)
602 		return (EBUSY);
603 
604 	/* Free any allocated resources */
605 	ppt_teardown_msi(ppt);
606 
607 	if (numvec == 0)		/* nothing more to do */
608 		return (0);
609 
610 	flags = RF_ACTIVE;
611 	msi_count = pci_msi_count(ppt->dev);
612 	if (msi_count == 0) {
613 		startrid = 0;		/* legacy interrupt */
614 		msi_count = 1;
615 		flags |= RF_SHAREABLE;
616 	} else
617 		startrid = 1;		/* MSI */
618 
619 	/*
620 	 * The device must be capable of supporting the number of vectors
621 	 * the guest wants to allocate.
622 	 */
623 	if (numvec > msi_count)
624 		return (EINVAL);
625 
626 	/*
627 	 * Make sure that we can allocate all the MSI vectors that are needed
628 	 * by the guest.
629 	 */
630 	if (startrid == 1) {
631 		tmp = numvec;
632 		error = pci_alloc_msi(ppt->dev, &tmp);
633 		if (error)
634 			return (error);
635 		else if (tmp != numvec) {
636 			pci_release_msi(ppt->dev);
637 			return (ENOSPC);
638 		} else {
639 			/* success */
640 		}
641 	}
642 
643 	ppt->msi.startrid = startrid;
644 
645 	/*
646 	 * Allocate the irq resource and attach it to the interrupt handler.
647 	 */
648 	for (i = 0; i < numvec; i++) {
649 		ppt->msi.num_msgs = i + 1;
650 		ppt->msi.cookie[i] = NULL;
651 
652 		rid = startrid + i;
653 		ppt->msi.res[i] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ,
654 							 &rid, flags);
655 		if (ppt->msi.res[i] == NULL)
656 			break;
657 
658 		ppt->msi.arg[i].pptdev = ppt;
659 		ppt->msi.arg[i].addr = addr;
660 		ppt->msi.arg[i].msg_data = msg + i;
661 
662 		error = bus_setup_intr(ppt->dev, ppt->msi.res[i],
663 				       INTR_TYPE_NET | INTR_MPSAFE,
664 				       pptintr, NULL, &ppt->msi.arg[i],
665 				       &ppt->msi.cookie[i]);
666 		if (error != 0)
667 			break;
668 	}
669 
670 	if (i < numvec) {
671 		ppt_teardown_msi(ppt);
672 		return (ENXIO);
673 	}
674 
675 	return (0);
676 }
677 
678 int
ppt_setup_msix(struct vm * vm,int bus,int slot,int func,int idx,uint64_t addr,uint64_t msg,uint32_t vector_control)679 ppt_setup_msix(struct vm *vm, int bus, int slot, int func,
680 	       int idx, uint64_t addr, uint64_t msg, uint32_t vector_control)
681 {
682 	struct pptdev *ppt;
683 	struct pci_devinfo *dinfo;
684 	int numvec, alloced, rid, error;
685 	size_t res_size, cookie_size, arg_size;
686 
687 	error = ppt_find(vm, bus, slot, func, &ppt);
688 	if (error)
689 		return (error);
690 
691 	/* Reject attempts to enable MSI-X while MSI is active. */
692 	if (ppt->msi.num_msgs != 0)
693 		return (EBUSY);
694 
695 	dinfo = device_get_ivars(ppt->dev);
696 	if (!dinfo)
697 		return (ENXIO);
698 
699 	/*
700 	 * First-time configuration:
701 	 * 	Allocate the MSI-X table
702 	 *	Allocate the IRQ resources
703 	 *	Set up some variables in ppt->msix
704 	 */
705 	if (ppt->msix.num_msgs == 0) {
706 		numvec = pci_msix_count(ppt->dev);
707 		if (numvec <= 0)
708 			return (EINVAL);
709 
710 		ppt->msix.startrid = 1;
711 		ppt->msix.num_msgs = numvec;
712 
713 		res_size = numvec * sizeof(ppt->msix.res[0]);
714 		cookie_size = numvec * sizeof(ppt->msix.cookie[0]);
715 		arg_size = numvec * sizeof(ppt->msix.arg[0]);
716 
717 		ppt->msix.res = malloc(res_size, M_PPTMSIX, M_WAITOK | M_ZERO);
718 		ppt->msix.cookie = malloc(cookie_size, M_PPTMSIX,
719 					  M_WAITOK | M_ZERO);
720 		ppt->msix.arg = malloc(arg_size, M_PPTMSIX, M_WAITOK | M_ZERO);
721 
722 		rid = dinfo->cfg.msix.msix_table_bar;
723 		ppt->msix.msix_table_res = bus_alloc_resource_any(ppt->dev,
724 					       SYS_RES_MEMORY, &rid, RF_ACTIVE);
725 
726 		if (ppt->msix.msix_table_res == NULL) {
727 			ppt_teardown_msix(ppt);
728 			return (ENOSPC);
729 		}
730 		ppt->msix.msix_table_rid = rid;
731 
732 		if (dinfo->cfg.msix.msix_table_bar !=
733 		    dinfo->cfg.msix.msix_pba_bar) {
734 			rid = dinfo->cfg.msix.msix_pba_bar;
735 			ppt->msix.msix_pba_res = bus_alloc_resource_any(
736 			    ppt->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
737 
738 			if (ppt->msix.msix_pba_res == NULL) {
739 				ppt_teardown_msix(ppt);
740 				return (ENOSPC);
741 			}
742 			ppt->msix.msix_pba_rid = rid;
743 		}
744 
745 		alloced = numvec;
746 		error = pci_alloc_msix(ppt->dev, &alloced);
747 		if (error || alloced != numvec) {
748 			ppt_teardown_msix(ppt);
749 			return (error == 0 ? ENOSPC: error);
750 		}
751 	}
752 
753 	if (idx >= ppt->msix.num_msgs)
754 		return (EINVAL);
755 
756 	if ((vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
757 		/* Tear down the IRQ if it's already set up */
758 		ppt_teardown_msix_intr(ppt, idx);
759 
760 		/* Allocate the IRQ resource */
761 		ppt->msix.cookie[idx] = NULL;
762 		rid = ppt->msix.startrid + idx;
763 		ppt->msix.res[idx] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ,
764 							    &rid, RF_ACTIVE);
765 		if (ppt->msix.res[idx] == NULL)
766 			return (ENXIO);
767 
768 		ppt->msix.arg[idx].pptdev = ppt;
769 		ppt->msix.arg[idx].addr = addr;
770 		ppt->msix.arg[idx].msg_data = msg;
771 
772 		/* Setup the MSI-X interrupt */
773 		error = bus_setup_intr(ppt->dev, ppt->msix.res[idx],
774 				       INTR_TYPE_NET | INTR_MPSAFE,
775 				       pptintr, NULL, &ppt->msix.arg[idx],
776 				       &ppt->msix.cookie[idx]);
777 
778 		if (error != 0) {
779 			bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, ppt->msix.res[idx]);
780 			ppt->msix.cookie[idx] = NULL;
781 			ppt->msix.res[idx] = NULL;
782 			return (ENXIO);
783 		}
784 	} else {
785 		/* Masked, tear it down if it's already been set up */
786 		ppt_teardown_msix_intr(ppt, idx);
787 	}
788 
789 	return (0);
790 }
791 
792 int
ppt_disable_msix(struct vm * vm,int bus,int slot,int func)793 ppt_disable_msix(struct vm *vm, int bus, int slot, int func)
794 {
795 	struct pptdev *ppt;
796 	int error;
797 
798 	error = ppt_find(vm, bus, slot, func, &ppt);
799 	if (error)
800 		return (error);
801 
802 	ppt_teardown_msix(ppt);
803 	return (0);
804 }
805